Google announced that the use of computers is now a built-in tool in Gemini 3.5 Flash, allowing developers to build AI agents that can interact across platforms such as browsers, desktops, and mobile environments. Previously only available as a standalone Gemini 2.5 computer-based model, this feature is now integrated directly into Gemini 3.5 flash.
Mateo Quiros, product manager for Google DeepMind, said the integration will deliver Google’s best-ever performance for agent-based computer tasks.
Gemini 3.5 Flash now available natively on your computer
Gemini already supports function calls and built-in tools like search and map grounding. Now that computer usage is built into the main Gemini Flash model, developers can use Gemini 3.5 Flash to reliably build custom agents that can see, infer, and perform actions across browser, mobile, and desktop environments.
Google said this feature improves performance for long-term enterprise automation tasks, such as continuous software testing and knowledge work across specialized applications.
Safety measures when using a computer
To reduce prompt injection risks for agents operating in live environments, Google uses targeted adversarial training for computational use in Gemini 3.5 Flash.
The company also introduces two optional enterprise protection systems that allow organizations to:
- Require explicit user confirmation before sensitive or irreversible actions.
- Automatically stops the task if an indirect prompt insertion attempt is detected.
Google said companies need to take a defense-in-depth approach, combining these safeguards with secure sandboxes, human verification, and strict access controls. Additional information regarding safety measures is available from the company’s best practices document.
availability
Developers and enterprises can start using their computers with Gemini 3.5 Flash through the Gemini API and Gemini Enterprise Agent Platform.
Google also provides a browser-based hosted demo environment for testing functionality. Developers can also access reference implementations and documentation to get started building agents for their computers.
